Back to skill

Security audit

New Player Package 800

Security checks for vulnerabilities and agentic risk

Overview

This is a setup guide rather than bundled malware, but it asks users to make broad, persistent OpenClaw changes through unpinned bulk installs and updates.

Install only in an isolated or non-critical OpenClaw profile after reviewing each named skill and package. Avoid running the one-click block or clawhub update --all blindly; pin versions where possible, verify publishers, and understand how to disable or remove task-persistence, filesystem, system-management, and memory/vector-search components before enabling them.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (1)

T08 · Insecure Dependencies

Warning
Location
SKILL.md:94
Finding
Unpinned and Unverified Third-Party Package and Skill Installation## Vulnerability Details **File Location**: `SKILL.md:94`, `SKILL.md:101-104`, and `SKILL.md:161` **Vulnerability Type**: Unverified and mutable third-party dependencies **Risk Level**: Medium ### Vulnerable Code ```bash # SKILL.md:94 npm install -g clawhub uv ``` ```bash # SKILL.md:101-104 # Install official skills clawhub install clawhub find-skills skill-creator clawddocs openclaw-anything clawdbot-filesystem ontology # Install enhanced skills clawhub install session-monitor task-persistence ``` ```markdown <!-- SKILL.md:161 --> 1. **Regular Updates**: `clawhub update --all` ``` ### Technical Analysis The deployment guide directs users to install global npm packages and nine additional OpenClaw Skills without pinning exact versions, immutable commit identifiers, or content hashes. It also does not specify canonical source URLs, verified publishers, package signatures, checksums, or a review procedure before activation. Consequently, the effective code and instructions installed by these commands can change after this Skill has been audited. The recommendation to run `clawhub update --all` further expands this trust boundary by accepting mutable releases of every installed Skill in bulk. Although the document describes some dependencies as official, it provides no locally verifiable provenance. An upstream account compromise, registry compromise, dependency-confusion condition, malicious package replacement, or unexpected future release could therefore introduce attacker-controlled behavior. The global npm installation may increase the affected host scope depending on the privileges and npm configuration used to execute it. ### Attack Path 1. An attacker compromises a publisher account, package registry entry, Skill distribution channel, or another upstream component associated with one of the named dependencies. 2. The attacker publishes a malicious package or Skill under a version that the unpinned installation or update command will select. 3. A ...[truncated 1534 chars]
Remediation
## Remediation Suggestions 1. **Pin immutable dependency versions** - Specify exact npm package versions rather than resolving the latest release. - Pin each Skill to an immutable version, commit identifier, or content digest where supported. - Store the reviewed dependency set in a lockfile or equivalent manifest. 2. **Verify package provenance** - Document canonical registry and repository URLs. - Identify expected publishers or organization accounts. - Require package signature, provenance attestation, or checksum verification before installation. - Publish expected hashes through a separate trusted channel. 3. **Avoid automatic bulk updates** - Replace `clawhub update --all` with individually reviewed upgrades. - Review release notes and source changes before updating each Skill. - Test updates in an isolated staging environment before production activation. 4. **Reduce installation privileges** - Avoid privileged or root execution for global npm installation. - Prefer a dedicated user account, project-local installation, container, or other isolated environment. - Restrict filesystem, network, credential, and system-management access to the minimum required by each Skill. 5. **Review Skills before activation** - Inspect each downloaded Skill's instructions, scripts, dependency declarations, and requested capabilities. - Reject Skills that retrieve mutable remote payloads or request permissions unrelated to their documented purpose. - Maintain an explicit allowlist of reviewed Skill names and pinned versions. 6. **Add rollback and monitoring controls** - Back up configuration and known-good Skill versions before upgrades. - Log package installation, Skill activation, and update events. - Monitor for unexpected filesystem, network, configuration, and persistence changes after installation.
Vulnerability Patterns
  • Rogue AgentSelf-Modification, Session Persistence
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (2)

Missing User Warnings

Medium
Confidence
93% confidence
Finding
The skill includes a 'one-click' script that performs system-modifying actions such as global package installation, creation of directories under the user's home directory, and bulk skill installation without any warning, confirmation step, or guidance to review the commands first. In an agent-skill context, this is dangerous because users may execute the block verbatim, causing unintended local system changes and implicitly trusting additional skills that may expand attack surface.

Session Persistence

Medium
Category
Rogue Agent
Content
npm install -g clawhub uv

# Clone and install core skills
mkdir -p ~/.openclaw/skills
cd ~/.openclaw/skills

# Install official skills
Confidence
76% confidence
Finding
The instructions recommend setting up persistent skill directories and installing multiple auxiliary skills into a long-lived location under ~/.openclaw/skills. Persistence itself is not automatically malicious, but in this context it creates durable changes to the environment and encourages continued trust in installed components, which can preserve unsafe behavior or broaden exposure if any installed skill is later found vulnerable or malicious.

Static analysis

No suspicious patterns detected.